The present invention relates to data management, and more specifically, to methods and systems for conditional write processing for a cache structure of a coupling facility.
In parallel computing systems, multiple computers or mainframes act together as a single system and share data using one or more coupling facilities. Data management middleware products running on the several systems participate in the data sharing environment by exploiting a cache structure, a lock structure and/or a list structure of the coupling facility. The cache structure allows the computing systems to store data in a high-performance cache that is accessible by all of the systems in the parallel computing system. The coupling facility maintains a shared directory that tracks the validity of individual data in each of the middleware local in-memory caches, which is used to maintain cache coherency and data integrity for the shared data across all middleware instances.
Certain kinds of write-intensive database activities such as batch jobs that modify most or all of the data items in a shared database can have the effect of overwhelming the cache structure of the coupling facility by flooding the structure with numerous changed data pages which are not part of the normal working set of the cache structure. For example, these activities may include database reorganizations, index restructurings, or making application-oriented updates across all data items. During such operations the cache structure of the coupling facility can become overwhelmed with a flood of write commands that displace the normal working set data in the cache structure, thereby causing temporary performance degradation or disruption and adversely affecting the performance of applications that are making use of the shared data.